Dipole density distribution for the δ band. AD indicates Alzheimer disease; LFD, left frontal δ; LOD, left occipital δ; LPD, left parietal δ; LPFD, left prefrontal δ; LTD, left temporal δ; MCI, mild cognitive impairment; RFD, right frontal δ; ROD, right occipital δ; RPD, right parietal δ; RPFD, right prefrontal δ; and RTD, right temporal δ.
Dipole density distribution for the θ band. AD indicates Alzheimer disease; LFT, left frontal θ; LOT, left occipital θ; LPFT, left prefrontal θ; LPT, left parietal θ; LTT, left temporal θ; MCI, mild cognitive impairment; RFT, right frontal θ; ROT, right occipital θ; RPFT, right prefrontal θ; RPT, right parietal θ; and RTT, right temporal θ.
Customize your JAMA Network experience by selecting one or more topics from the list below.
Fernández A, Turrero A, Zuluaga P, et al. Magnetoencephalographic Parietal δ Dipole Density in Mild Cognitive Impairment: Preliminary Results of a Method to Estimate the Risk of Developing Alzheimer Disease. Arch Neurol. 2006;63(3):427–430. doi:https://doi.org/10.1001/archneur.63.3.427
Subjects with mild cognitive impairment (MCI) are at a higher risk of experiencing Alzheimer disease (AD). Magnetoencephalographic temporoparietal dipole densities of low-frequency activity are good predictors of individuals' cognitive status, and might be a useful tool to investigate the conversion from MCI to AD.
To investigate the role of low-frequency dipole densities as predictors of the risk of developing AD.
Whole-head magnetoencephalographic recordings were obtained from 19 probable AD patients, 17 MCI patients, and 17 healthy control subjects. The generators of focal magnetic slow waves were located using a single moving dipole model.
Left parietal δ dipole density permitted a reliable classification of AD and MCI patients. The MCI patients were divided into 2 groups based on the median left parietal δ dipole density, and were followed up for 2 years. The estimated relative risk of conversion to AD was increased by 350% in those MCI patients with high left parietal δ dipole density scores.
Results confirmed the important role of parietal δ dipole density in the evaluation of AD and MCI. A magnetoencephalographic-based assessment of AD and MCI patients might be considered a useful clinical test in the near future.
Alzheimer disease (AD) is the most frequent cause of dementia, and emerging evidence suggests that structural and functional changes may occur 20 to 30 years before the onset of clinical symptoms.1 The long duration of a preclinical or “prodromal” phase of AD has substantially increased efforts devoted to determine its main features and the factors associated with developing the full dementia syndrome. Mild cognitive impairment (MCI) is one of the terms used to describe the transitional state between normality and dementia. Patients with MCI are at a higher risk of experiencing AD, with a conversion rate from MCI to dementia of 10% to 15% per year, while healthy control subjects convert at a 1% to 2% rate.2 (A more detailed description can be found in the review by Shah et al.3)
A quantitative electroencephalographic investigation presented promising results for early detection of conversion to AD. For example, a progressive reduction of θ power is associated with cognitive decline.4 α and θ relative power in left temporo-occipital derivation correctly classifies 85% of MCI subjects who eventually develop AD.5 An alternative method for electroencephalographic analysis, global field power and dipole source calculation, reveals that a more anterior localization of θ and α activity was the best predictor of future development of AD.6 Dipole source calculation by magnetoencephalography (MEG) has also been used for AD research. The calculation of dipole density (DD) for low-frequency (δ and θ) bands demonstrated significantly increased temporoparietal δ activity in AD patients.7 Moreover, these studies indicated that temporoparietal δ DD predicted the variability of an individual's cognitive status. The present study provides preliminary evidence of the efficacy of parietotemporal δ activity, as measured by MEG, to predict MCI-to-AD conversion. In addition, a model to discriminate AD from MCI was created.
Fifty-three subjects recruited from the Geriatric Unit, Hospital Universitario San Carlos de Madrid, Madrid, participated in the study: 19 probable AD patients, 17 MCI patients, and 17 healthy controls. Clinical diagnosis of AD was made following medical, neurological, psychiatric, and neuropsychological examinations. Patients with AD fulfilled the criteria for probable AD,8 and the MCI diagnosis followed the criteria of Petersen et al.9 All of the MCI patients were followed up for a 2-year period. Cognitive status was screened in all groups with the Spanish version of the Mini-Mental State Examination,10 while functional status was evaluated by the GDS/FAST system.11Table 1 summarizes the demographic and cognitive/functional status of the subjects.
Patients and controls were free of significant medical, neurological, and psychiatric diseases (other than AD or MCI), and were not using drugs that could significantly affect MEG activity. Before the MEG recording, all subjects or legal representatives signed an informed consent that explained the technical and ethical considerations of the technique. The study was approved by the local ethics committee.
The MEG was measured using a 148-channel whole-head magnetometer (MAGNES 2500 WH; 4D Neuroimaging, San Diego, Calif) during a 5-minute resting period. Recordings were obtained in a lying condition, and subjects were asked to stay awake and to avoid head and eye movement by fixating on a black point on the ceiling. The MEG was recorded with a 678.17-Hz sampling rate, using a bandpass filter of 0.1 to 200 Hz. Eye movements (electro-oculograms) were recorded from 4 electrodes attached to the left and right outer canthus and above and below the left eye. The electrocardiogram was monitored with electrodes attached to the right collarbone and the lowest left rib. Each 5-minute data set was decimated (which consisted of filtering the data to respect the Nyquist criteria,12 followed by downsampling by a factor of 16) and bandpass filtered before the analysis in the δ (1.5- to 4.0-Hz) and θ (4.0- to 8.0-Hz) bands. The DDs were estimated based on an equivalent current dipole in a homogeneous sphere calculated for each time step.
Artifact-free time segments of varying lengths were determined by visual inspection. Single equivalent dipoles were fitted for each time point in the selected epochs. The location of activity sources was computed with reference to a cartesian coordinate system defined by a set of 3 anatomical landmarks (fiduciary points): the right and left external meatus and the nasion. The position of the magnetometers relative to the subject's head was precisely determined using 5 coils, of which 3 were attached to the fiduciary points and 2 to the forehead. During the recording session, a fiberoptic motion detector was used to ensure that the subject's head did not change position relative to the sensor.
Only dipole-fit solutions at time points with a root mean square over 30 m and a goodness of fit over 0.90 were accepted for further analysis. For statistical analysis, the total brain volume was divided into 10 regions, representing frontal, parietal, prefrontal, temporal, and occipital areas in both hemispheres, and the number of successful dipole fits was determined for each of these regions. The total number of dipoles per second fitting the criteria in the δ and θ bands was determined for each subject and region (Figure 1 and Figure 2, respectively). To avoid the possible influence of the individual time segment lengths over the DD scores, the raw number of dipoles per region was normalized by calculating the following formula:
where the absolute number of recorded points is the final number of recorded points after eliminating the artifacts.
This final score was used for the statistical analysis of the resulting 20 DD variables: 2 sides (left and right) × 5 regions (frontal, parietal, parietal-frontal, temporal, and occipital) × 2 frequency bands (δ and θ). This method has been broadly described elsewhere.7,13-17
A polytomous logistic regression was performed on the dependent variable “diagnosis,” which was divided into 3 categories: AD, MCI, and control. Two nonredundant logit functions were calculated using MCI as the reference category: logit 1 for AD-MCI comparison and logit 2 for control-MCI comparison. The variable selection process began with a careful univariate analysis of each DD variable, using the likelihood ratio test as the selection criterion. Seven variables (left parietal δ DD [LPD], right parietal δ DD, left temporal δ, right occipital δ [ROD], left parietal θ, right parietal θ, and left temporal θ) demonstrated a predictive power in the univariate analysis. Among these variables, a multivariate stepwise procedure only selected LPD (likelihood ratio test, P<.001) as the significant factor for a final polytomous model. However, the regression coefficients were significant for the logit 1 function (P = .002), but not for the logit 2 function (P = .11), indicating that LPD would only permit a good classification of AD vs MCI patients, but not of MCI patients vs controls. We then tested the remaining significant variables from the univariate analyses, and found that ROD was significant for the logit 2 function (P = .05), but not for the logit 1 function (P = .95). Thus, there were 2 individual binary logistic regression models: AD vs MCI and control vs MCI.
Once the assumption of linearity in the logit scale was checked and accepted, the LPD-based model was called model 1, and some of its corresponding statistics are as follows: when the variable was constant, the estimated coefficients for the constant term and the variable LPD were 1.42 and −1.50, respectively. The P value of the Hosmer-Lemeshow goodness-of-fit test was P = .29, and the area under the receiver operating characteristic curve was 0.81 (95% confidence interval, 0.66-0.95). The observed vs predicted responses for model 1 when a 0.5 cut point was adopted were as follows: when the observed diagnosis was MCI, MCI was the predicted diagnosis in 13 patients and AD was the predicted diagnosis in 4 patients, leading to a specificity of 76%; when the observed diagnosis was AD, AD was the predicted diagnosis in 14 patients and MCI was the predicted diagnosis in 5 patients, leading to a sensitivity of 74%. The total classification accuracy was 75% (27 of 36 patients).
The addition of the variable ROD × ln(ROD) to the model containing ROD yielded a significant P value (P = .01). The ROD × ln(ROD)–based model was called model 2, and its corresponding statistics are as follows: when the variable was constant, the estimated coefficient for the constant term was 0.59; the estimated coefficients for ROD and ROD × ln(ROD) were 3.09 and 8.41, respectively. The P value of the Hosmer-Lemeshow goodness-of-fit test was P = .20, and the area under the receiver operating characteristic curve was 0.80 (95% confidence interval, 0.65-0.95). The observed vs predicted responses for model 2 when a 0.5 cut point was adopted were as follows: when the observed diagnosis was MCI, MCI was the predicted diagnosis in 12 subjects and control was the predicted response in 5 subjects, leading to a sensitivity of 71%; when the observed response was control, MCI was the predicted response in 2 subjects and control was the predicted response in 15 subjects, leading to a specificity of 88%. The total classification accuracy was 79% (27 of 34 patients).
The 17 MCI patients were followed up for 2 years, with a clinical examination every 6 months; no loss to follow-up occurred. The cohort was divided into 2 groups based on a median split of the LPD measure: MCI high and MCI low.
The 2 groups of MCI patients did not differ in terms of age or Mini-Mental State Examination score at baseline. After 2 years, 5 (29%) of 17 patients met the criteria for probable AD. The Mini-Mental State Examination scores were significantly (P = .04) reduced (mean ± SD, 8.80 ± 7.01 points) in the group of converters to AD from baseline. Of the 8 patients in the MCI high group, 4 (50%) developed AD, while only 1 (11%) of 9 patients in the MCI low group converted to dementia (Table 2) (relative risk, 4.5; 95% confidence interval, 0.9-28.3).
This preliminary study replicates prior findings that AD patients have increased δ DD in parietotemporal regions.7,13,14 More important, however, we provide evidence that high LPD scores are a valid marker of risk to convert from MCI to AD within 2 years. Elevated δ activity can be explained by the interaction of 2 processes: deafferentation and reduced cholinergic levels. Corticothalamic and corticocortical deafferentation are major sources of δ activity, and both are present in AD patients' brains.18,19 Furthermore, there is a significant correlation between loss of cholinergic neurons in the nucleus basalis, reduced acetylcholinesterase levels, and δ power in AD patients. This correlation was further supported by quantitative electroencephalographic and quantitative MEG studies in which scopolamine (a cholinergic antagonist) infusions generated changes in background activity, inducing an increased δ and θ power.20,21
Mild cognitive impairment is a complex concept, and a recent review by Petersen2 suggests 2 models. In the first model, cognitive decline from normal aging to AD consists of a progressive reduction of an individual's functioning, and limits between normal aging, MCI, and probable AD can be well-defined. The second model considers MCI as interposed between cognitive changes of normal aging and those that might constitute early dementia. The key aspect of this interpretation resides in the degree of overlap between normal aging and MCI, but especially between MCI and early dementia. Such overlap prevents a clear definition of the transition between MCI and AD.
Our results clearly support the second perspective. The logistic regression analyses showed some degree of overlap, represented by incorrect classification of the MCI patients. Obviously, while these errors could represent a limitation of the test, it is important to note that 3 (75%) of 4 of the MCI patients incorrectly classified as having AD actually went on to develop AD. This raises the question of where the border is between MCI and AD. Morris et al22 studied 3 groups of MCI patients with differing degrees of cognitive impairment. All participants were enrolled in a longitudinal follow-up study, and at final examination, all MCI patients had progressed to a more severe dementia. When the MCI patients' brains were examined, 21 (84%) of 25 had histological signs of AD, 2 had signs of vascular dementia, and 1 had frontotemporal dementia. They concluded that “individuals considered by current criteria to have only MCI in fact have very mild AD.”22(p402)
Early detection of those MCI patients at higher risk of conversion to AD is increasingly becoming a fundamental need clinically and for research purposes. The results of our investigation may have a potentially important clinical value. Medications currently used (ie, acetylcholinesterase inhibitors) for AD treatment delay the process of deterioration, and may have a better effect when cholinergic functioning is still not too low. Early detection will be even more critical when primary prevention therapies reach the clinic. In the meantime, use of MEG may help us to understand the physiological relationship between MCI and AD, and the factors that may accelerate the transition to AD.
Correspondence: Tomás Ortiz, MD, PhD, Centro MEG Dr Pérez Modrego, Facultad de Medicina, Pabellón 8, Universidad Complutense de Madrid, 28040 Madrid, Spain (firstname.lastname@example.org).
Accepted for Publication: October 10, 2005.
Author Contributions:Study concept and design: Fernández, Gil, Campo, and Ortiz. Acquisition of data: Maestú and Campo. Analysis and interpretation of data: Fernández, Turrero, Zuluaga, and Gil. Drafting of the manuscript: Fernández and Ortiz. Critical revision of the manuscript for important intellectual content: Turrero, Zuluaga, Gil, Maestú, and Campo. Statistical analysis: Turrero and Zuluaga. Obtained funding: Ortiz. Study supervision: Fernández, Gil, Maestú, and Ortiz.
Acknowledgments: This study was funded by Fundación Pérez-Modrego and project SEJ2004-06969 of the Spanish Ministerio de Educación y Ciencia, both located in Madrid. We thank M. Santiuste, MD, PhD; J. Martín, MD; J. T. Becker, PhD; R. Yubero, MA; and L. Morón, MA, for their invaluable collaboration.
Create a personal account or sign in to: